Asset Prepayment Assumptions

RECORD OF SOCIETY

1995 VOL. 21 NO. 2

ASSET

Moderator:

Panelists:

Recorder:

PREPAYMENT

OF ACTUARIES

ASSUMPTIONS

STEPHEN D. REDDY

STEVE W. ABRAHAMS*

RANDALL L. BOUSHEK

CATHERINE E. EHRLICH

STEPHEN D. REDDY

This session will address asset prepayments from several perspectives. Presentations will

cover emerging prepayment experience in the mortgage-backed security arena, special

modeling considerations when addressing prepayments, and the initial experience and

impact of the NAIC flow uncertainty index (FLUX).

MR. STEPHEN D. REDDY: Asset prepayment assumptions and asset prepayments

themselves are a very important part of modeling assets and projecting future cash flows.

It's important to understand both what drives prepayments and also the total effect of

prepayments on an organization. We will cover asset prepayments from a couple different

angles. One will be a discussion of the NAIC FLUX model, which is a regulatory screening tool for measuring cash flow uncertainty and variability. Second will be a discussion of

what drives prepayments, particularly mortgage collateral, and modeling considerations

relating to those prepayment assumptions.

Our first speaker will be Randy Boushek, a vice president and portfolio manager at

Lutheran Brotherhood, which is a fraternal benefit society headquartered in Minneapolis,

with approximately $15 billion under management. His responsibilities include overseeing

all trading, research and portfolio management within the life company bond portfolio,

serving as Investment Division liaison on asset/liability management issues, directing

quantitative investment research, and overseeing derivative activities across all fixed

income portfolios. He's a member of the NAIC's technical resource groups on collateralized mortgage obligation (CMO) accounting and CMO cash flow volatility that developed

the FLUX model. He's a frequent speaker at both actuarial and investment conferences.

Second, we have Steve Abrahams, who's a vice president at Morgan Stanley & Company.

Steve is a member of the mortgage research group in the fixed income division and spends

a great deal of time studying mortgage prepayments in the residential sector. I think he has

some interesting things to say about what's been happening in prepayments the last couple

years and what Morgan Stanley is doing in its modeling efforts with respect to

prepayments.

Third, we have Catherine Ehrlich, who's a senior vice president at Capital Management

Sciences (CMS), which provides fixed income software systems and consulting services to

the investment management community. As a manager in the New York office, Catherine

is responsible for marketing and client support on the East Coast. Prior to joining CMS,

she was an assistant vice president with Metropolitan Life Insurance Company, where she

*Mr.Abrahams,nota memberof thesponsoringorganizations,isVicePresidentof MorganStanley,& Co.

inNew York,NY.

261

RECORD, VOLUME 21

had various assignments including cash-flow testing, pricing, and customer support for

institutional pension products. She earned her bachelor of arts degree at Colgate University

and is a Fellow of the Society of Actuaries and a Chartered Financial Analyst.

I'm Steve Reddy, also with Morgan Stanley. I work in the portfolio strategies group in the

Fixed Income Division in New York, doing asset/liability consulting, primarily for life

companies. I've had the pleasure of working with Steve. We're part of the same group, but

he's much more of an expert in the mortgage prepayment arena.

Randy will lead offand talk about the FLUX system, and then we'll get into some of the

other issues involving mortgage prepayments and modeling considerations relating to that.

MR. RANDALL L. BOUSHEK: My assignment on the panel is to discuss the NAIC's

FLUX model and to tie that discussion to the general topic of prepayment assumptions. As

Steve mentioned, I am a member of the technical resource group that developed the FLUX

model under the auspices and at the direction of the NAIC's invested asset working group

(IAWG).

My outline for this presentation consists of six points. First, I would like to discuss in

general what the FLUX model is and, perhaps, more importantly, what it is not. Second, I

want to provide a brief history of the model's development and discuss some of the

considerations that led it to take its current form. Third, and most germane to this session, I

want to focus on the scenario specification and prepayment assumptions that are necessarily

a part of the FLUX model. Fourth, and with a promise to avoid all formulas, I want to

discuss the functional mechanics of the model. Fiffia, I'd like to review with you a distribution of actual FLUX scores for 1994. Finally, I'd like to comment briefly on one or two

open issues.

Just what is FLUX? The FLUX model is a regulatory screening tool. It was developed

specifically for insurance regulators for the sole purpose of enabling them to narrow the

multitude of cash-flow testing reports that they receive to a more manageable few that may

require closer scrutiny because of potential CMO cash-flow volatility. The FLUX model is

specifically not a rating mechanism, nor a tool for establishing reserving requirements or

pass/fail tests. It is also not a portfolio management tool. To clear up a bit of confusion, it

should also be emphasized that FLUX is a calculation model, not a prepayment model. It

does use and set the specification for a set of prepayment assumptions provided by the

Public Securities Association (PSA), but it is neither a prepayment model nor a valuation

model of any kind.

The technical resource group which developed the model was comprised of representatives

from Wall Street, the insurance industry, and investment software vendors who provided

much of the horsepower for testing various designs and aspects of the model. Our general

charge from the IAWG was to develop a methodology for assessing the relative cash flow

volatility of individual CMO tranches, with three or four specific constraints. Two key

words here are relative and individual. The FLUX model provides a relative measure of

volatility exposure--there

is no absolute interpretation to any FLUX score. Further, the

FLUX model only attempts to quantify the potential for cash flow volatility without

assessing whether that volatility would be good or bad for the holder. More contentiously,

262

ASSET PREPAYMENT ASSUMPTIONS

the model is specifically designed to evaluate bonds on a trancbe-by-tranche (as opposed to

aggregate-portfolio) basis, at the direction of the IAWG.

There were three other specific guidelines provided by the IAWG. First, the model was to

have an "open architecture," meaning that all formulas were to be publicly available and

reproducible by any interested party, and all assumptions were to be under control of the

regulators. No "black box" approaches were acceptable. Second, the model was to produce

a single score for each CMO wanche across all companies, that is, a score that is independent of book value. Certainly, ifI own a bond at a price of 80, and you own the same bond

at a price of 120, we have very different statutory risk exposures. Nonetheless, given its

specific objective, the model was expressly developed to be independent of holding price.

Finally, the model had to be as simple as possible.

As to the development process itself, we initially considered three different models submitted by members of the technical resource group. The FLUX model in its original form was

developed and submitted by Andrew Davidson. It underwent several refinements over the

course of more than a year before it was submitted in final form to the IAWG. Beyond the

design of the model itself, several other issues had to be addressed: How many scenarios,

and which ones? How do we combine results from various scenarios? What prepayment

assumptions do we use? How do we accommodate floating rate instruments? I'll address

scenarios and prepayment assumptions in a moment. The question of combination is

actually quite interesting. The FLUX model produces a single numerical value, or score,

for each CMO tranche for each scenario. Given a vector of scores corresponding to a set of

scenarios for a given tranche, one might make an argument (and more than one did) for

either the maximum or mean as the best representative score for the bond. In the end, after

considerable deliberation, we settled on root mean square as the most acceptable

compromise.

The specification of scenarios and prepayment assumptions is designed to be a dynamic

process, under the control of insurance regulators. Striking a balance between simplicity

and a representative range of outcomes is particularly difficult here. Based in part on

spanning set research by Dr. Thomas Ho, an initial set of five nonlevel interest rate

scenarios, plus a base case level rate scenario, was established for the test year 1993. These

five scenarios included two increasing rate paths, two decreasing rate paths, and one

interest rate whipsaw. With the benefit of further research, a second whipsaw scenario was

added for 1994, and the oscillation period of the earlier whipsaw was compressed. The

addition, deletion, or alteration of scenarios in the future is entirely at the discretion of

regulators.

Once the scenarios have been established and a freeze date is set to determine the initial

level of interest rates, dealers are surveyed fortheir prepayment projections as of the freeze

date for a three-dimensional array (agency, program, and coupon) of pass-through collateral

for the specified scenarios. All such projections are provided as a vector of monthly

prepayment assumptions for each scenario for each collateral cell in the matrix. Once this

information has been received, a median speed is calculated for each month in each vector.

The resulting matrix of vectors of monthly median speeds then becomes the "official" set of

prepayment assumptions for the FLUX model. At any point in time, any broker, vendor, or

insurer can calculate "current" FLUX scores using their own proprietary prepayment

263

RECORD, VOLUME 21

assumptions; however, the NAIC's "annual" FLUX scores are based solely on the official

set of prepayment assumptions, which are publicly available to any interested party.

While the FLUX model utilizes an "official" set of prepayment assumptions, there is

technically no "official" FLUX score for any bond. Given the prepayment vectors determined above, and the discount rate and volatility assumptions which rll discuss in a

moment, any broker, vendor, or insurer with an accurate CMO structuring model should be

able to produce the same cash flows and FLUX scores for any CMO tranche. Annual

scores are submitted to the NAIC electronically by broker dealers and distributed to

regulators via the state data network. For simplicity, the bulk of the scores are submitted

by Merrill Lynch via its Passport system, with scores for bonds not modeled in Passport

submitted by other brokers. Life companies are not responsible for either calculating or

reporting FLUX scores.

Regulators can electronically match a company's Schedule D CMO holdings to the master

list of FLUX scores on the state data network for their own analysis. Since this is the

limited purpose that the FLUX model was designed for, FLUX scores are not published in

the insurer's financial statements.

Beyond prepayments, the FLUX model requires two other input assumptions: discount rate

and volatility. These assumptions are also set annually by the NAIC. The discount rate is,

as one might expect, an interest rate used to discount cash flows in the calculation model.

However, the volatility assumption is a specific variable in a few of the FLUX model

calculations and not the more typical controlling input into a stochastic interest rate

generator or options model. The technical resource group did develop a formulaic

approach for determining these assumptions, but as promised earlier I will not get into any

specifics here. Suffice it to say that on the basis of this approach the discount rate for 1993

and 1994 was set at 6% and 7.50%, respectively, while the volatility rate was set at 1.75%

for 1993 and 2% for 1994.

The FLUX score for any given bond for any given scenario is the sum of two components--a present value measure and a timing measure. The present value measure reflects

the magnitude of negative percentage change in present value in each scenario relative to

the base case. The timing measure represents the sum of period-by-period differences in

scaled cumulative present value of cash flows in each scenario relative to the base case.

The present value measure is designed to capture the risk of adverse prepayments on the

valuation of bonds priced at a significant premium or discount to par, for example, interest

only (IOs) and principal only (POs). The timing measure is designed to capture the risk of

adverse prepayments on the reinvestment of cash flows, particularly for companion/support

and "jump" tranches. Two key words in these definitions are negative and absolute, and

they underscore a key point about the FLUX model. Namely, that it assesses only the

potential volatility of cash flows without regard to direction (shortening or extending) and

with no offset for beneficial impact. This also helps explain why the sum of the parts, that

is, the weighted average FLUX score for each tranche in a CMO or in the simplest case for

an IO and a PO, almost always exceeds the whole, or the FLUX score, for the underlying

collateral.

264

ASSET PREPAYMENT ASSUMPTIONS

Table 1 shows the distribution of annual FLUX scores for 1994 for the CMO universe as

defined by Bloomberg, excluding those bonds for which no FLUX score was submitted.

The information presented here includes FLUX scores for approximately 33,000 tranches.

The first line shows the distribution of FLUX scores by number of tranches, while the

second line is the distribution by dollar value. On a dollar-weighted basis, over 50% of the

tranches in the universe had a FLUX score below three at year-end 1994, while almost 75%

were below five. In general, 1994 scores are lower than 1993 scores, owing to changes in

the starting level of interest rates, projected prepayment assumptions, and the discount rate

and volatility assumptions. For the sake of comparison, I have also included the FLUX

score for a current coupon Government National Mortgage Association (GNMA)

pass-through, which is actually higher than the score for more than half of the dollarweighted CMO universe. It is interesting to note that the same GNMA pass-through had a

FLUX score in excess of six for 1993.

TABLE 1

DISTRIBUTIONOF 1994 FLUX SCORES

0-1

1-3

3-5

5-10

10-20

> 20

Average

CMO Universe

37%

22%

12%

11%

7%

11%

6.6

CMO Universe

(S-weighted)

18%

35%

21%

14%

6%

6%

5.3

?

?

?

?

?

?

?

Life Co Holdings

(S-weighted)

FLUX Scorefor 30-year GNMA 8% Pass-Through:3.6

Source:Bloomberg

LP5/04_95(universeincludes33,000 tranches)

It is important to remember that "high" FLUX scores are not necessarily bad. High scores

indicate the potential for cash-flow volatility. This is not the same as indicating a high

level of risk. A concentration of high scores may conceal offsetting risks within the

portfolio, or ignore offsetting risks or liability exposures outside of the portfolio. High

FLUX scores are simply a cue to regulators that a closer look at actuarial cash-flow testing

results may be warranted.

I'd like to comment on one or two other items before closing my remarks. First, I mentioned earlier that "current" FLUX scores can be computed at any time by any broker,

vendor or insurer. Beyond the obvious difference in prepayment assumptions, these scores

will vary from "annual" FLUX scores for two other reasons (even if the annual prepayment

vectors are used). For one thing, the starting point for interest rate scenarios will in all

likelihood be different, and for another, some principal may have already paid down since

the freeze date for the annual scores. As a caveat, current scores are not necessarily

indicative at any point in time of subsequent annual scores.

Second, I also mentioned earlier that FLUX was specifically designed as a bond-by-bond

tool. While it is theoretically possible to calculate a portfolio FLUX score by aggregating

all projected cash flows before calculating the timing and present value measures, the

usefulness of such a number is debatable. Beyond the desire of regulators to be able to

independently assess the potential cash-flow volatility of individual investments, the

265

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download